The Comprehensive Upper Air Observation Network (CUON) Dataset

The Copernicus Climate Change Service (C3S) has developed the Comprehensive Upper Air Observation Network (CUON) dataset. The main geophysical variables included in CUON are temperature, humidity, and wind. The input observation data are the NOAA Integrated Global Radiosonde Archive (IGRA), the NCAR Upper-Air Database (UADB), the ERA5 observation feedback archive, and additional ascents from smaller collections, including in particular the African Monsoon Multidisciplinary Analysis (AMMA) and the World Ozone and Ultraviolet Radiation Data Centre (WOUDC). Available radiosonde, ozonesonde, and pilot-balloon (PILOT) platforms are included, even if the station record contains only a single launch. Key improvements over the aforementioned data input are the following: balloon drift estimates, observation error estimates and homogeneity adjustments for the main variables. The actual launch times were also refined as far as possible from the nominal times of reporting plus available metadata (e.g., IGRA release times). These unique features make CUON particularly suitable as an input for climate reanalysis, in particular the upcoming ERA6 reanalysis, but also other climate applications.

Comparison with ERA5 gridded data shows a sizeable reduction of representation errors and biases across all main variables, in particular in the early 2000s but also at other time periods back to the 1940s. The offline calculated observation minus background (obs-bg) departures are sometimes 30% smaller than those calculated during ERA5 assimilation. This may be explained by the offset of radiosondes during ascent as compared to their launch position, which can reach several 100km, i.e. several reanalysis gridboxes. 

The obs-bg departures form the basis for comprehensive statistics-based adjustment of biases in temperature, wind direction and also humidity, using the RAOBCORE/RICH method. The corresponding software has been further improved compared to the past year, with a better treatment of data gaps. 

Results from bias-adjusted temperature records indicate realistic spatial trend heterogeneity and a better fit to reprocessed satellite data products than what could be achieved in preparation to the present operational reanalysis ERA5. Temperature background departures from ERA5 increase substantially, both in terms of mean and standard deviations when going back to the early 1950s and 1940s. The present investigated whether this increase comes from poorer quality observations or from issues arising due to the less strongly observationally constrained ERA5 state during this period.

Humidity bias adjustments prove to be more delicate to implement, since it is not sufficient to shift the distributions by a mean value. Instead, it turns out to be important to adjust also the shape of the distributions. To achieved this, a quantile matching approach has been adopted, taking into account the size of the change of background departures in the time intervals before and after a potential breakpoint. The adjustment led to a reduction of obs-bg departures with respect to ERA5 but also reduced the strong spurious drying trends over continental-scale networks such as the US and China in the period 1990-2020.

The CUON dataset goes back to 1905 and will be updated at least annually. It will be made available from: https://cds.climate.copernicus.eu/datasets/insitu-comprehensive-upper-air-observation-network?tab=overview.